Starship Build Engineering System

Verify

← Back to outcome

VERIFY — Starship Build Engineering System

How verification runs

node verify.mjs imports the engines directly from src/, runs 18 assertions, and writes verification-report.json (machine) and verification-report.md (human) plus a copy under evidence/. Exit code is non-zero if any check fails.

What each check asserts

#CheckWhat it proves
1Takt = 480/8 = 60takt formula
2Theoretical min stations = ceil(500/60) = 9work-content / takt bound
34×30-min chain @ takt 60 → 2 stations, 100% effbalancer packs tasks optimally in a known case
4Vehicle balance respects precedence, station ≤ takt, min ≤ N ≤ opsno illegal assignment
5Line-efficiency identity = work/(stations·cycle), 0 < eff ≤ 1metric is internally consistent
6Balancing is deterministicsame input → same layout
7Throughput = 480/60 = 8bottleneck-limited rate
8Parallel stations for 130 min @ takt 60 = 3parallelization sizing
9CPM A(3),B(4)→C(2) → 6, critical B,Cforward/backward pass on a tiny known net
10Vehicle flow → 56 h, critical A,B,E,F,G, slacks D=14 C=2full CPM on the model
11Cp/Cpk of [9,10,11] vs 7..13 = 1.0/1.0capability indices
12Φ(0)=0.5, Φ(1.96)≈0.975, symmetry, erf(0)≈0normal-CDF accuracy
13±3σ → yield ≈ 0.9973, ≈ 2700 PPMfallout model
14tol [0.1,0.2,0.2] → WC 0.5, RSS 0.3exact stack-up
15RSS ≤ worst-case for every vehicle stackstatistical bound invariant
1640→12 parts = 70%, 100→40 s = 60%DFM reduction math
17DFA: 2 essential parts, 2 candidatesconsolidation logic
18Balanced line ≤ stations and ≥ efficiency vs baselinebenchmark sanity

Checks 1–17 are data-independent (they use fixtures with known answers), so they keep guarding the math if you swap the notional model for real data. Check 18 is the synthetic benchmark on the model.

What verification does NOT assert

It does not assert real-world accuracy, real production rates, or that any real weld/fit/part figure is correct — the model is notional (see proof/LIMITATIONS.md).